Murphy, Geoff Grant2025-12-092025-12-092025https://hdl.handle.net/10566/21552The 21cm hyperfine transition of neutral hydrogen, HI, is an invaluable source of cosmological information, with its detection offering the opportunity to constrain models of the Universe, and thereby improving our understanding of its evolution. Numerous, and ever more sensitive, experiments target this signal across a range of frequencies, and, correspondingly, a range of redshifts, from Cosmic Dawn, to the Dark Ages, the Epoch of Reionisation, and the present-day Universe. Common among these experiments, however, are numerous performance-limiting effects, be it signal chain systematics, foreground contamination, etc. This work focuses on the mitigation of these examples in two different 21 cm experiments — signal chain systematics in simulated Hydrogen Epoch of Reionisation Array (HERA) data, and the separation of foregrounds from the 21 cm signal in the MeerKAT Large Area Synoptic Survey (MeerKLASS), both in a statistically rigorous Bayesian framework. The first chapter statistically models signal chain systematics in simulated HERA visibilities, namely cable reflections and antenna cross-coupling. We employ a Hamiltonian Monte Carlo (HMC) sampler to constrain our forward-model of the systematics-corrupted autocorrelation and cross-correlation visibilities as observed in HERA, and use our results to remove the systematics from mock data. This method allows us to not only recover the underlying power spectra, but to also provide rigorously propagated statistical uncertainties, a property which is important in the reporting of upper limits on the Epoch of Reionisation (EoR) signal. In the presence of high noise, we are able to effectively mitigate the systematics down to the noise level. Incoherent averaging of the recovered visibilities further reduces the noise, suggesting there is minimal residual systematics present. In cases of negligible noise, the majority of the systematics can be mitigated, but a fair amount of residual systematics remain, primarily due to the density of the systematics, i.e. confusion between systematics. However, this work performs similarly in terms of recovery in comparison to the currently implemented filtering and f itting methods in the HERA observational pipeline, but with the addition of statistical uncertainty estimates being placed on the recovered visibilities and power spectra. There is no significant signal loss in any of our results, an important consideration in the recovery of the weak Hi signal. The second chapter separates the 21 cm signal from the orders-of-magnitude more powerful foregrounds in simulated single-dish intensity maps, with future applications to MeerKLASS (MeerKAT Large Area Synoptic Survey) data being the intended outcome of this work.en21cmHeraMeerklassBayesian modelling approachesMeerkatThesis